Blog
Mastering Tomorrow's Tech Today
Tutorials
Yes, you can train your own LLM from scratch if you think small. I trained a 1.5M parameter model on a 2-core machine with no GPU and watched it learn in 25 minutes; a 64M MiniMind chatbot costs under a dollar of rented GPU time.
A ChatGPT summary writes a new paragraph; TextFold folds the file you already have into five depths. I ran TextFold on a…
Read More
Rust already links with lld on x86-64 Linux, so most people do not need mold anymore. I timed GNU ld, lld, mold…
Read More
A self hosted Alexa alternative is five small programs in a row: wake word, speech to text, intent, action and text to…
Read More
tsrs vs tsc on one TypeScript folder: both printed the same 81 errors byte for byte, tsrs was about 1.7 times faster,…
Read More
Can AI predict stock prices? On 30 real Bitcoin forecasts, the open source Kronos model missed by about 3% a day ahead,…
Read More
DeepSeek published the program that writes DeepSeek-V4.1-Flash's replies, not the warehouse that makes a cluster serve 155 trillion tokens a day. The…
Read More
Microsoft did not ship a smarter model on 19 October. It described a badge that starts empty, a risk score that moves…
Read More
Google TV's own settings cannot remove the sponsored banner. Here is how to switch off ad, tracking and preloaded apps over ADB,…
Read More
One person aimed AI agents at company websites and got into more than 100 of them for about USD 5,000 in tokens.…
Read More
harness2 does not need an API key for its local demo. I ran it with no key and a 0.0 cost field,…
Read More